How Mathematics Improves The World | Measuring Inequality Without Looking at Only One Person’s Income
Suppose one household earns $4,000 a month.
Is the society unequal?
You cannot know.
If every household earns $4,000, the distribution is perfectly equal by income.
If half earn $4,000 and half earn $40,000, inequality is substantial.
If almost everyone earns around $4,000 and one household earns $4 million, another shape appears.
Inequality is not a fact about one person.
It is a fact about relationships across a distribution.
That makes measurement harder than asking for an average.
Which incomes count?
Before or after taxes?
Do government transfers count?
Should a household of six be compared directly with a household of one?
Do we measure income or wealth?
Individuals or households?
One year or lifetime resources?
Mathematics does not make those definitional choices disappear.
It makes their consequences visible.
Quick Read
The OECD defines income inequality as the difference in how income is distributed across a population and measures it using cumulative population shares against cumulative income shares. The World Bank describes the same geometry through the Lorenz curve and defines the Gini index as the area between that curve and the line of perfect equality, scaled by the maximum possible area.
A Gini coefficient of 0 represents perfect equality under the chosen income concept. A coefficient approaching 1 represents extreme concentration. But the Gini is not “the inequality”. Different distributions can share the same Gini while differing greatly at the top, bottom or middle. Economists therefore also use income shares, percentile ratios, Palma ratios, Theil indices, Atkinson indices and other measures.
Singapore’s Department of Statistics provides a useful current example of why definitions matter. In Key Household Income Trends, 2025, released in February 2026, the Gini coefficient based on household market income per household member among resident households was 0.452 before Government transfers and taxes. After accounting for Government transfers and taxes, it was 0.379. The report identifies both as the lowest values in their respective series since the current household-market-income series began in 2015. The 2025 data are preliminary.
The point is not that one number proves a society is fair or unfair. The point is that a distribution can be measured consistently, compared through time and decomposed under different definitions. Mathematics gives public debate a common object to examine.
One-sentence answer: Mathematics improves the world by converting thousands or millions of individual incomes into transparent measures of distribution—revealing concentration, change and redistribution patterns that no single salary, average or anecdote could show by itself.
Start With the Distribution, Not the Average
Society A has incomes:
10, 10, 10, 10, 10.
Mean = 10.
Society B:
0, 0, 0, 0, 50.
Mean = 10.
Same average.
Radically different distribution.
An average answers “how much in total per person on average?”
Inequality asks “how is that total spread?”
Median Helps, But Still Does Not Describe the Shape
The median is the middle observation after sorting.
It resists extreme values better than the mean.
But two societies can share the same median and have different tails.
Income distributions require multiple summaries because location and spread are different properties.
Sort Everyone From Lowest Income to Highest
The Lorenz curve begins by sorting the population by income.
Then calculate cumulative shares.
Bottom 10% of people receive what share of income?
Bottom 20%?
Bottom 50%?
Bottom 90%?
Plot cumulative population on the horizontal axis and cumulative income on the vertical.
A distribution becomes geometry.
The Line of Perfect Equality
If every person receives the same income, the bottom 20% receives 20% of total income.
The bottom 50% receives 50%.
The Lorenz curve is the diagonal:
L(p)=p.
Real income distributions lie below this line because lower-income groups receive less than their population share of total income.
The more bowed the curve, the greater the concentration under this representation.
The Gini Coefficient Is an Area
Let A be the area between the line of equality and the Lorenz curve.
Let A+B be the total area below the equality line.
Then:
G = A/(A+B).
Because A+B is 1/2 when both axes run from 0 to 1:
G = 1 − 2∫₀¹L(p)dp.
The World Bank definition describes the same geometric idea, usually reporting the index on a 0–100 scale.
The Gini turns a whole curve into one number.
A Pairwise Interpretation
For non-negative incomes, Gini can also be expressed using average absolute pairwise income differences.
G = [1/(2n²μ)] ΣiΣj|yi−yj|
where μ is mean income.
This interpretation is intuitive.
Choose pairs of people.
Measure how different their incomes are.
Normalise by mean income.
Inequality becomes average social distance in income space.
Why the Gini Is Popular
It is scale invariant.
Double every income and the Gini remains unchanged.
It uses the whole distribution rather than one cutoff.
It is comparable through time when definitions and data quality are consistent.
It has clear geometric and pairwise interpretations.
Those strengths made it a standard international indicator.
Why One Gini Is Not Enough
Two different Lorenz curves can produce the same area.
One society may have severe concentration at the top and a relatively compressed lower 90%.
Another may have wide differences throughout the middle.
Same Gini.
Different lived distribution.
A scalar summary necessarily discards shape information.
Lorenz Curves Can Cross
Distribution A gives the poorest 20% more income than Distribution B.
But B gives the bottom 70% more cumulatively.
The Lorenz curves cross.
No simple Lorenz-dominance ordering exists.
Different inequality indices can then rank the distributions differently because they weight parts of the distribution differently.
Disagreement between metrics can be mathematically legitimate.
Percentile Ratios: Compare Concrete Places in the Distribution
P90/P10 compares income near the 90th percentile with income near the 10th.
P50/P10 compares median with the lower tail.
P90/P50 compares upper middle with median.
These measures are easy to interpret.
They ignore what happens beyond the selected percentiles.
Every simplicity chooses what not to see.
Income Shares: Ask Who Receives the Total
What share of national or household-survey income goes to:
- bottom 10%?
- bottom 50%?
- top 10%?
- top 1%?
Income shares describe concentration directly.
They are especially useful when researchers care about the tails rather than the whole distribution equally.
The Palma Ratio: Focus on the Tails
The Palma ratio compares the income share of the richest 10% with that of the poorest 40%.
Its motivation is that middle-income shares can be relatively stable across many settings while tail shares vary more.
A Palma ratio of 1 means the top 10% receives the same aggregate income as the bottom 40%.
It highlights a particular distributional contrast rather than summarising every pairwise difference.
Theil Index: Inequality as Entropy
The Theil T index can be written:
T = (1/n) Σ (yi/μ) ln(yi/μ).
It comes from information theory.
One major advantage is decomposability.
Total inequality can be separated into within-group and between-group components under appropriate grouping.
This helps analysts ask:
How much inequality comes from differences between regions, and how much exists inside regions?
Atkinson Index: Put an Ethical Parameter Into the Formula Explicitly
The Atkinson family includes an inequality-aversion parameter ε.
Higher ε gives greater weight to income differences near the bottom.
This makes a normally hidden normative choice visible.
There is no value-free way to collapse a distribution into one social-welfare-equivalent number.
The Atkinson index says so mathematically.
Income and Wealth Are Not the Same Distribution
Income is a flow over time.
Wealth is a stock of assets minus liabilities at a point in time.
A retired homeowner may have modest current income and substantial wealth.
A young professional may have high income and negative net wealth because of debt.
Income inequality cannot be casually described as wealth inequality.
The numerator changed meaning.
Market Income and Disposable Income Answer Different Questions
Market income reflects earnings and other market-related sources under a stated definition.
Disposable-income concepts subtract direct taxes and add cash transfers depending on statistical system.
Some national analyses also estimate in-kind benefits.
Comparing before and after transfers and taxes measures the distributional effect of the fiscal system under that accounting framework.
The two Ginis are not competing estimates of one identical quantity.
They describe different income concepts.
Singapore 2025: One Distribution, Two Policy Stages
Singapore’s Department of Statistics reported in February 2026 that the Gini coefficient based on household market income per household member was 0.452 in 2025 before Government transfers and taxes.
After accounting for Government transfers and taxes, the coefficient was 0.379.
The report states that both were the lowest since records under the current household-market-income series began in 2015.
It also marks 2025 figures as preliminary.
The difference between 0.452 and 0.379 is not a claim that all policy effects are captured perfectly.
It is a defined statistical comparison under the report’s transfer-and-tax accounting method.
Household Size Changes Interpretation
A household earning $8,000 with one person and a household earning $8,000 with six people have the same total household income.
Their material resources per person differ.
Divide by household size and you obtain per-member income.
But simple per-capita division assumes no economies of scale.
Housing and utilities are partly shared.
Equivalence scales try to adjust for household composition more subtly.
Equivalence Scales: A Family of Denominators
One common approach divides household income by the square root of household size.
Another uses modified OECD weights for the first adult, additional adults and children.
Singapore’s statistical publications show Gini coefficients under several equivalence approaches in methodological comparisons.
Different scales answer slightly different welfare questions.
The denominator is part of the model.
Household and Individual Inequality Are Not Interchangeable
Two individuals with different wages may live in the same household and share resources.
A household-based measure treats them differently from an individual earnings distribution.
Wage inequality.
Individual income inequality.
Household market income inequality.
Disposable household income inequality.
These should be named precisely before comparisons are made.
Zero and Negative Incomes Complicate Some Indices
Some inequality formulas involve logarithms or ratios.
Zero or negative incomes can make them undefined or hard to interpret.
Survey statisticians therefore define treatment rules carefully.
Gini itself can also require care when negative values occur.
Real data does not always behave like textbook positive numbers.
Survey Data Has Sampling Error
National inequality measures often come from household surveys.
The survey observes a sample, not every household.
Weights expand observations to represent the population.
Sampling design creates uncertainty.
Bootstrap or replication methods can estimate standard errors for Gini and quantile statistics.
A change from 0.381 to 0.379 may be numerically real in the estimate and still require inference before declaring a meaningful structural shift.
Nonresponse Is Not Random Automatically
Very high-income households may be harder to survey.
Very low-income or unstable households can also be underrepresented.
If missingness correlates with income, ordinary survey weights may not eliminate bias completely.
Administrative tax records can complement surveys, especially near the top.
Different data sources see different parts of the distribution well.
Top Coding and Tail Measurement
Datasets may cap reported incomes for privacy or robustness.
If all incomes above a threshold are recorded at that threshold, top concentration is understated.
Researchers may use Pareto-tail models or administrative data to improve top-income estimates.
Measurement of inequality is especially sensitive to what happens in the tails.
Price Levels Matter Across Countries
A $2,000 income does not buy the same basket everywhere.
Purchasing-power adjustments help compare absolute living standards across countries.
Within-country Gini coefficients are invariant to multiplying everyone’s income by the same exchange-rate factor.
But cross-country welfare comparisons need more than Gini.
Equality and prosperity are separate axes.
A Perfectly Equal Poor Society Can Have Gini Zero
Everyone earns 1.
Gini = 0.
Everyone earns 100.
Gini = 0.
The Gini says nothing about the absolute income level when everybody is scaled equally.
GDP per capita, median income, poverty rates and inequality measures answer different questions.
A serious dashboard needs more than one axis.
Growth Incidence: Who Benefited From Growth?
Median income rises 5%.
Gini falls.
Useful.
But which percentiles gained?
A growth-incidence curve plots income growth rates across quantiles.
The bottom may grow faster than the top.
Or all groups may grow while the top grows fastest.
Distributional change is a vector across society, not only a before-and-after coefficient.
Mobility: Cross-Sectional Inequality Is Not Lifetime Inequality
Students often have low current income and higher future income.
Retirees may move from high earnings to lower retirement income.
If people move substantially through the distribution over a lifetime, annual cross-sectional inequality can differ from lifetime-resource inequality.
Panel data follows the same households through time.
Mobility matrices ask:
what fraction of the bottom quintile moves upward after five years?
Static inequality and mobility are distinct dimensions.
Between-Group and Within-Group Inequality
Split a population by region.
Some inequality comes from different regional means.
Much may remain within every region.
Decomposable measures such as Theil allow analysts to separate these components.
This prevents a common mistake:
seeing group differences and assuming they explain the entire national distribution.
Counterfactual Decomposition
Inequality changed.
Why?
Age structure changed.
Employment patterns changed.
Returns to education changed.
Household composition changed.
Taxes or transfers changed.
Decomposition methods construct counterfactual distributions to estimate contributions.
But decomposition is model-dependent.
Explanation requires causal care beyond arithmetic differences.
Redistribution Measurement Is Not a Moral Verdict
If the post-transfer Gini is lower than the pre-transfer Gini, the measured fiscal system reduced inequality under those definitions.
That does not by itself tell us:
- whether tax rates are optimal;
- whether benefits create other effects;
- whether public services are efficient;
- whether opportunity is equal;
- whether the resulting distribution is socially preferred.
Measurement establishes a distributional fact.
Policy evaluation needs additional objectives and evidence.
Income Inequality Is Not Opportunity Inequality
Two people can earn different incomes because of choices, age, hours, skill, luck, discrimination, inherited advantage or many interacting causes.
A Gini cannot separate them.
Opportunity measures need information about circumstances and outcomes across generations or groups.
Outcome inequality and opportunity inequality are not synonyms.
Income Inequality Is Not Poverty
Everyone’s income can double.
Gini remains unchanged.
Poverty can fall sharply.
Or inequality can fall because top incomes decline while low incomes do not improve.
Poverty asks whether resources are below a threshold.
Inequality asks how resources are distributed.
Both matter.
Neither substitutes for the other.
Income Inequality Is Not Consumption Inequality
Households can smooth consumption through saving and borrowing.
A temporary income fall may not reduce consumption equally.
Some countries measure welfare using consumption expenditure because income is harder to observe reliably.
The World Bank’s Gini metadata explicitly notes that the underlying welfare concept may be income or consumption depending on the dataset.
Cross-country comparisons should check the concept first.
Data Definition Can Move the Number Without Society Changing
A statistical agency expands household coverage.
Changes treatment of government contributions.
Revises survey weights.
The reported series can shift even if underlying behaviour did not suddenly change that year.
Good time-series analysis reads methodology notes before reading the chart.
Continuity of definition is part of comparability.
International Comparisons Need Harmonisation
Country A reports disposable household income per equivalent adult.
Country B reports market income per capita.
Comparing the two Ginis as if they were identical variables is invalid.
OECD and World Bank datasets work to harmonise concepts, but users still need to inspect metadata.
Comparability is a mathematical and institutional achievement.
Uncertainty Can Be More Than Sampling Error
Measurement error in income.
Nonresponse.
Underreported capital income.
Informal earnings.
Household composition assumptions.
All create uncertainty beyond ordinary sampling variance.
Sensitivity analysis recalculates inequality under alternative plausible assumptions.
A robust conclusion survives reasonable measurement choices.
A Dashboard Beats a Single Number
For a serious picture of household economic distribution, combine:
- median income;
- income by decile;
- Gini;
- top and bottom shares;
- poverty or low-income measures where defined;
- mobility data;
- wealth where available;
- before/after transfer comparisons.
No single indicator is a civilisation score.
Multiple measures preserve more of the distribution.
A Classroom Thought Experiment: Same Mean, Different Gini
Group A incomes:
10, 10, 10, 10, 10.
Group B:
2, 4, 8, 12, 24.
Both total 50.
Both mean 10.
Sort each group.
Compute cumulative income shares.
Draw Lorenz curves.
The difference becomes visible immediately.
A Second Thought Experiment: One Transfer
Start with incomes:
5, 10, 15, 20, 50.
Transfer 5 from the richest person to the poorest without changing total income:
10, 10, 15, 20, 45.
Mean stays 20.
The Lorenz curve moves closer to equality.
Gini falls.
Students see how a distribution can change while the average does not.
Primary Mathematics: Inequality Begins With Sorting and Fractions
Primary students already know:
- ordering numbers;
- fractions;
- percentages;
- averages;
- bar charts;
- cumulative totals.
Sort five incomes.
Find how much the bottom two receive.
Compare with their 40% population share.
The Lorenz curve begins in Primary arithmetic.
Secondary Mathematics: Distribution Becomes Geometry and Statistics
Secondary students add:
- percentiles;
- cumulative frequency;
- area under curves;
- ratios;
- variance;
- sampling.
Lorenz curves become cumulative graphs.
Gini becomes area.
Percentile ratios become direct comparisons.
Sampling error reminds us the national number is estimated from data.
Advanced Mathematics: Inequality as Distributional Statistics
Modern distribution analysis draws on:
- measure and probability theory;
- survey sampling;
- quantile statistics;
- information theory;
- welfare economics;
- decomposition methods;
- causal inference;
- panel-data analysis.
The Gini is only the entry point.
The deeper subject is how a distribution changes, why it changes and what a chosen summary keeps or discards.
Why This Improves the World
1. It replaces anecdote with distribution
One salary cannot describe society; distributional statistics can.
2. It makes change measurable through time
Consistent definitions allow governments, researchers and citizens to examine whether concentration is rising or falling.
3. It shows what taxes and transfers change
Before/after comparisons quantify redistribution under explicit accounting rules.
4. It prevents one metric from becoming the whole story
Lorenz curves, percentile ratios and alternative indices reveal features a single Gini hides.
5. It makes definitions inspectable
Household size, income concept, survey coverage and equivalence scales can be stated rather than silently assumed.
6. It improves public reasoning
People can disagree about desired distributions while still sharing a transparent measurement of the distribution they are debating.
What Mathematics Does Not Do
A Gini coefficient does not say whether a society is just.
It does not measure absolute prosperity.
It does not measure wealth unless wealth is the variable used.
It does not prove why inequality changed.
It does not measure opportunity automatically.
It does not make different national definitions comparable by magic.
And no single inequality number can preserve the entire shape, history and human meaning of an economic distribution.
Frequently Asked Questions
What is the Gini coefficient?
The Gini coefficient is a summary measure of distributional inequality derived from the Lorenz curve. Zero represents perfect equality under the chosen variable and population definition; higher values represent greater concentration.
What is a Lorenz curve?
A Lorenz curve plots cumulative population share, ordered from lowest to highest income, against the cumulative share of total income received by that population.
Can two countries have the same Gini and different inequality patterns?
Yes. The Gini compresses the whole distribution into one area measure, so different Lorenz-curve shapes can produce the same coefficient.
Why do Gini figures differ before and after taxes and transfers?
They measure different income concepts. Transfers add resources to some households and taxes subtract them, changing the distribution and therefore the Gini.
What was Singapore’s household-income Gini in 2025?
Singapore DOS reported a preliminary 2025 Gini of 0.452 for household market income per household member before Government transfers and taxes, and 0.379 after accounting for Government transfers and taxes under the report’s definitions.
Is lower inequality always better?
That is a normative and policy question, not something a Gini coefficient can answer alone. Evaluation also considers living standards, mobility, incentives, poverty, opportunity, public services and other social objectives.
Sources and Further Reading
- OECD, Income Inequality, defining the indicator and Gini coefficient using cumulative population and income shares.
- World Bank DataBank, Gini Index Metadata, explaining the Lorenz curve and 0–100 Gini index.
- Singapore Department of Statistics, Key Household Income Trends, 2025, released February 2026, including household market-income Gini measures before and after Government transfers and taxes.
- Singapore Department of Statistics, Household Income, current data, publications and methodological resources.
Continue Through eduKateSG
Continue with How Mathematics Works. Compare this article with Dividing Something Fairly When Everyone Values It Differently: one measures an observed distribution while the other studies rules for creating allocations. It also connects to Ranking Teams That Never All Play Each Other, because both show how a single ordering or scalar can hide the structure of the evidence beneath it.
Final Thought: Inequality Is the Shape Between People
One household tells you its income.
Another does the same.
Millions more follow.
Sort them.
Add their shares.
Draw the curve.
Measure the area.
Then refuse to stop there.
Look at the top.
The bottom.
The middle.
Before policy.
After policy.
Across time.
Mathematics improves the world here by turning a political, economic and human argument into something that can at least begin with a distribution everyone is able to inspect.